AI-Driven Fuzzing for Vulnerability Assessment of 5G Traffic Steering Algorithms

TL;DR

This paper tackles vulnerability assessment of 5G Traffic Steering (TS) algorithms under adversarial conditions by proposing an AI-driven fuzzing framework built on NSGA-II. The framework integrates with the Sionna-based simulator to generate adversarial network states and optimize a three-objective fitness: $f_1(z)$ (instability), $f_2(z)$ (QoE degradation), and $f_3(z)$ (unfairness). Across six scenarios and five TS algorithms, AI-Fuzzing detects $8{,}207$ vulnerabilities versus $6{,}112$ with traditional testing, including $200$ critical failures compared to $189$, and shows faster convergence and greater diversity (Shannon index $0.924$ versus $0.777$). The results support adopting AI-driven fuzzing as a primary, architecture-aware validation method for robust 5G/6G TS systems, while noting the stochastic nature of rare vulnerabilities and outlining future work in real-world testbeds and GPU-accelerated fuzzing.

Abstract

Traffic Steering (TS) dynamically allocates user traffic across cells to enhance Quality of Experience (QoE), load balance, and spectrum efficiency in 5G networks. However, TS algorithms remain vulnerable to adversarial conditions such as interference spikes, handover storms, and localized outages. To address this, an AI-driven fuzz testing framework based on the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) is proposed to systematically expose hidden vulnerabilities. Using NVIDIA Sionna, five TS algorithms are evaluated across six scenarios. Results show that AI-driven fuzzing detects 34.3% more total vulnerabilities and 5.8% more critical failures than traditional testing, achieving superior diversity and edge-case discovery. The observed variance in critical failure detection underscores the stochastic nature of rare vulnerabilities. These findings demonstrate that AI-driven fuzzing offers an effective and scalable validation approach for improving TS algorithm robustness and ensuring resilient 6G-ready networks.

Figures (2)

• Figure 1: Comparison of AI-Fuzzing and traditional testing across 300 runs per method (6 scenarios $\times$ 5 algorithms $\times$ 10 trials): AI-Fuzzing detects 34.3% more vulnerabilities per run (27.36$\pm$8.60 vs. 20.37$\pm$10.98, $p<0.00001$; $n=300$, t-test) with strong effect size (Cohen's d=0.708) (a). Convergence analysis shows AI-Fuzzing reaches 90% of optimal detection rate 25% faster than traditional testing (generation 9 vs. 12) (b).
• Figure 2: Vulnerability severity distribution across 300 runs (6 scenarios $\times$ 5 algorithms $\times$ 10 trials): AI-Fuzzing detects more total vulnerabilities (8,207 vs. 6,112, $p < 0.00001$) and critical failures (200 vs. 189, $p = 0.002$ Mann-Whitney U test) (a) . Normalized distribution shows superior diversity (Shannon: 0.924 vs. 0.777) with balanced severity coverage (b). High variance in critical failures (SD $>$ mean) reflects stochastic discovery of rare vulnerabilities.